Closed michael-schwarz closed 4 months ago
For protection, the problem seems to be that both get added to P
when it should in fact be empty.
In the set
, it seems like we only join the cpa
components after a non-definite assign and forget to also join priv
components. :thinking:
During #1456 I discovered that, in fact, all of our non-relational privatizations except the one casting Miné's analsysis in our framework are unsound.
The problem is that upon setting
*ptr = 5
, all targets in the may-point-to-set are updated, rather than joining together the result of updating the different targets. This is likely due to some bot/top confusion inside the local states of the base analysis where the non-present bindings are all assumed to be \bot.